On 3/5/10 10:39 PM, Alexey Proskuryakov wrote:
On 05.03.2010, at 10:27, Olli Pettay wrote:
I was going to mention this as the primary reason why frame bytes should
be included. JavaScript code needs this information for flow control,
Why?
I assume you are asking why JavaScript code needs this information for
flow control.
My recollection is that feature was added as a result of discussions
about implementing flow control. How else are you supposed to know that
you're streaming too fast without modifying the server? Since WebSockets
is a match for TCP/IP, and the latter provides ways to adaptively change
data rate, it's natural that one expects the same from WebSockets.
And you're saying that javascript really needs to know about the
frame boundary bytes to detect if it is streaming too fast.
Doesn't sound likely to me.
and it's raw bytes that are sent over the tubes, not original message
strings.
Right. But this is about the API. I assume the underlying protocol may
change or the API can eventually support different kinds of protocols
(some may use UPD, some TCP, some send text, some binary).
The API usage should be still the same, if possible.
This is something we agree about.
I guess the root of our disagreement is in how one uses the API. I'm
saying that the interesting question is how many bytes are buffered to
be sent over the wire, so in order to keep the API usage the same we
need to include protocol overhead in this number.
Or we need not to include protocol overhead in this number.
I'd like bufferedAmount to work the same way as .loaded and .total in
XHR+progress events; protocol specific headers aren't counted in.
In WebKit, we'd have to queue
unsent messages separately just to implement this quirk (see
https://bugs.webkit.org/attachment.cgi?id=50093 for a proof of concept).
It becomes very difficult to implement we decide to add size of data
that an underlying network library buffers internally - which I think
would be a reasonable thing to do.
I don't see why that would be difficult. If you know you have just
written x bytes to the whatever network method, you know how many bytes
of those were frame markers.
That's true, but I don't know how many of these have already been sent
unless I perform lots of additional bookkeeping.
"lots"
Consider sending "data"
message three times in a row:
\x00data\xFF\x00data\xFF\x00data\xFF
If we are to exclude protocol overhead in bufferedAmount, and we know
that there are 8 bytes still queued (a\xFF\x00data\xFF), and we know
that there were three frames sent (with an overhead of 6 bytes) how
would we know that the answer is 5?
You could for example have a circular list for the frame boundary byte
indexes and when something is removed from the queue, you just update
the pointer which points to the first frame boundary byte index in the
queue. Then if the circular list is implemented in a reasonable way,
counting all the frame boundary bytes in the queue is just one
subtraction.
-Olli
- WBR, Alexey Proskuryakov
